TEACHER WORKSHOP

Think & Speak

AI ENGLISH LEARNING PLATFORM

A deep dive into the technology that powers personalized vocabulary training,
AI-driven pronunciation coaching, and intelligent learning analytics for modern classrooms.

Click anywhere or press → to begin the workshop
01

The Learning Ecosystem at a Glance

One platform. One unified experience. From assessment to mastery.

📝

Assessment

8-dimension profiling captures age, proficiency, interests, motivation, and learning style to personalize every student's journey from the first click.

📚

Course Adventure

Structured learning paths with 5-task lesson structure: Video → Quiz → Glossary → Download → Speaking. Progress tracked with milestone achievements.

🎯

Vocabulary Trainer

5-stage mastery pipeline from pronunciation to dictation. Spaced repetition algorithms combat the forgetting curve. Words become flashcards automatically.

🎭

Conversation Role Play

Structured academic speaking with AI feedback. Phoneme-level pronunciation analysis identifies specific sound improvements. Scenarios from presentations to debates.

🌐

Live Translation

Real-time English ↔ Cantonese/Mandarin classroom support. Bilingual subtitles, smart note capture, automatic flashcard generation from lesson content.

📊

Analytics Dashboard

Teacher dashboard with real-time engagement heatmap. Track progress across modules. Identify struggling students before frustration builds. Data-driven instruction.

🔗

Key Insight: These are not separate tools — they form an integrated ecosystem. Assessment results drive course recommendations. Speaking practice feeds vocabulary lists. Every interaction trains the AI. Listen → Capture → Review → Master.

02

The Assessment Engine

Personalized learning starts with understanding the learner

How It Works

1. Age & Learning Stage

Select from groups: 6-8, 9-11, 12-14, 15-17, 18+. Adapts interface complexity and content delivery speed.

2. English Proficiency

1-10 scale with detailed descriptors for accurate placement. Colors adapt per age: gold for 6-8, teal for 9-11, blue for 12-14, orange for 15-17.

3. Interests & Topics

11 categories including Space & Science, Animals & Nature, Sports & Games, Technology. Powers personalized content selection across all modules.

4. Learning Style

8 methods: Play games, Read stories, Speak & act, Watch videos, and more. The AI matches teaching methodology to preferred style.

Technical Implementation

STATE Redux Profile Management

  • The selectUser Redux selector extracts tokenStudentPlatform for authentication
  • Assessment data dispatches via appStudentPlatformActions.setState
  • Profile persists across sessions via localStorage or secure cookies
  • useLatest hook prevents stale closures during async operations

INTEGRATION How Assessment Drives Everything

  • Vocabulary difficulty auto-adjusts based on proficiency level
  • Conversation Role Play scenarios match selected interests
  • AI Coach learns from 8-dimension profile to personalize hints
  • Dashboard adapts display based on age group and learning goals
💡

Teacher Insight: This is NOT a test — students don't fail. The system learns preferences to deliver content that fits each learner.

ONBOARDING Progressive Disclosure

The mascot character guides students through assessment with speech bubbles — white rounded containers with directional tails pointing to the mascot. Generous padding accommodates CJK character sets (Mandarin/Cantonese). First-time users see elaborate animations; returning users skip directly to content.

03

9 Pedagogical Interaction Types

Every learning modality covered — from watching to producing

🎬

Video Tasks

Multi-resolution streaming (1080P adaptive). Multilingual subtitle tracks (English, 繁體中文, 簡體中文). Transcript objects with timestamped note capture.

📝

Text Tasks

Structured reading with rich text formatting. Comprehension exercises designed for diverse reading abilities. Contextual vocabulary highlighting.

Quiz Tasks

Single-choice (MC) and multiple-select (MCS) formats. Time limits, explanations, immediate validation. isAllQuestionsAnswered gating.

🎤

Speaking Tasks

AI pronunciation analysis with phoneme-level feedback. Azure Speech Services integration. Four-dimensional scoring: Fluency, Accuracy, Completeness, Prosody.

🔀

Drag & Drop

Matching and sequencing activities. TaskContentDragAndDropQuestion and TaskContentDragAndDropAnswer interfaces. Kinesthetic learning support.

🔲

Fill in the Blank

Interactive written exercises with immediate validation. Vocabulary and grammar practice. AI-tolerance for spelling variations beyond exact matching.

📤

Upload Tasks

Project submissions and assignment uploads. Portfolio-based learning. Teachers can assign differentiated materials to student subgroups.

🌍

External Tasks

API key and callback URL for third-party integration. YouTube, TED talks, interactive grammar exercises — all within the platform's secure iframe sandbox.

⬇️

Download Tasks

Offline resources for homework. PDF viewer with iPad compatibility (pdfjs-dist Canvas API). Activity logging tracks every download interaction.

04

5-Stage Vocabulary Mastery Pipeline

From receptive understanding → analytical processing → expressive production

🎤 Stage 1

Pronunciation

✏️ Stage 2

Spelling I

📖 Stage 3

Usage I

🔀 Stage 4

Usage II

✍️ Stage 5

Dictation

🎤 Stage 1 — WordPronunciation: Foundation Layer

Students listen to and repeat target words. AI analyzes phoneme-level accuracy using Azure Speech Services. This stage establishes the phonetic memory foundation — if students can't say it correctly, they can't use it fluently.

  • Maximum 5 attempts per word (prevents frustration-based abandonment)
  • 3-second "think time" before recording begins (reduces anxiety)
  • Silence detection auto-terminates recording after 1 second of quiet

✏️ Stages 2 — Spelling I & II: Orthographic Layer

Students identify correct spellings within sentence contexts rather than in isolation. This implements contextual orthography — students learn words as they appear in real reading, improving recall in authentic communication.

📖 Stages 3 & 4 — Word Usage: Syntactic Application

SentencePickingPractices configuration supports usage-based learning — students select correct contextual applications rather than memorizing definitions. Stage 3 uses controlled contexts; Stage 4 requires advanced discrimination between similar words.

✍️ Stage 5 — Dictation: Expressive Production

The capstone task: students listen to audio and type the word from memory. This tests the deepest level of vocabulary retention — orthographic recall without visual cues. Represents the full receptive-to-expressive gradient.

🧠

Spaced Repetition Built In: Words not yet mastered automatically surface for review. The WrongWordsList routes struggling students to review before retrying. This implements Ebbinghaus forgetting curve combat — right when students are about to forget, the system brings words back.

📊

5 Dimensions Tracked: Vocab Guide, Spelling Basic, Spelling Advanced, Vocab Usage, Dictation. Radar chart visualization in the dashboard shows students their strengths and weaknesses at a glance.

05

AI-Powered Speaking Assessment

Phoneme-level precision — not just right or wrong, but exactly what to fix

Four-Dimensional Scoring

🎯
85%
Accuracy Score

Phoneme-by-phoneme precision. Identifies exactly which sounds the student mispronounced. Compare /θ/ vs /s/, /æ/ vs /ɛ/.

🎵
78%
Fluency Score

Prosodic timing and rhythm. Measures pauses, hesitations, and natural speech flow. Important for sounding like a native speaker.

92%
Completeness

Did the student say every part of the word? Low completeness indicates rushing or difficulty with the full pronunciation.

🔊
75%
Prosody Score

Intonation patterns — pitch variation, stress, and overall naturalness. Critical for tonal language learners adapting to English stress patterns.

How It Works

🎙️ Audio Capture Pipeline

Two implementation paths ensure browser compatibility:

  • MediaRecorder (modern): Adaptive MIME type selection (WebM/Opus vs WAV)
  • RecordRTC (legacy): StereoAudioRecorder at 16kHz for speech API optimization
  • Both systems support blob generation for backend upload

🤖 AI Assessment Engine

callGetTaskPracticeResult API retrieves assessment data. The system parses pronResultObj JSON responses containing phoneme-by-phoneme accuracy scores. Filters for valid assessment data excluding recording errors.

🎯

Key Differentiator: Other apps say "Wrong." We say "Your /θ/ needs work — try touching your tongue to your teeth." That's the power of phoneme-level feedback.

📋 Attempt Management

  • Maximum 5 attempts per exercise prevents brute-forcing
  • isAbleToPracticeAgain disables button when limit reached
  • WrongWordsList routes students to review before retrying
  • Tracking via callUpdateTaskWithUserData persists progress
06

Live Translation: Break Language Barriers

Real-time English ↔ Cantonese/Mandarin — turn every class into a learning ecosystem

How It Works for Teachers

📋 Create

Set up room

🎤 Teach

Speak normally

📊 Monitor

See engagement

✅ Review

Access analytics

🌐 Trilingual Architecture

  • English (en-US), Traditional Chinese 繁體中文 (zh-HK/zh-Hant), Simplified Chinese 简体中文 (zh-CN)
  • Real-time bidirectional: English ↔ Cantonese ↔ Mandarin
  • Font stack includes Noto Sans CJK SC/TC for optimal Chinese rendering
  • Line-height 1.6-1.8 for CJK glyph vertical complexity

🔄 WebSocket Communication

  • Separate channels: Chat (translations) vs Room (presence updates) — translations never blocked by student join/leave
  • 10-second refresh for room metadata and attendance
  • Auto-reconnect with exponential backoff (max 3 attempts)
  • Translation caching reduces latency for common phrases

Smart Capture Pipeline

🔊 Listen

Real-time bilingual subtitles appear as teacher speaks. Azure Speech Services captures phoneme-level data simultaneously.

📝 Capture

Students select important text → Add personal remarks → Categorize by topic. Unicode-aware regex extracts vocabulary automatically.

🎴 Review

Captured words auto-convert to flashcards. Spaced repetition brings them back at optimal review intervals. Teachers never manually create vocab lists.

🏆 Master

Students track progress through 5-stage vocabulary pipeline. "Every class becomes a structured, reusable learning experience."

07

Dashboard & Analytics

Data-driven teaching — see what your students need before they ask

47
Event Types Tracked
5s
Analytics Latency
200+
ML Features
<200ms
AI Response Time

What Teachers See

📈 Engagement Heatmap

Real-time visualization of student engagement: Green = Active, Yellow = Distracted, Red = Struggling. Teachers identify problem areas without interrupting the lesson. See who needs help at a glance.

🔔 Intervention Signals

When a student clicks the "Question" button (IconQuestions), teachers see an alert on the dashboard. The AI Coach correlates help-seeking behavior with performance data — frequent help requests + low scores = intervention trigger.

📊 Progress Tracking

useCallGetUserDashboard fetches real-time data. Tracks: Study Days, Study Minutes, Words Mastered, Spoken Times. Color-coded change metrics (green for positive, red for declining) show trends at a glance.

The Analytics Engine

📡 Activity Logging

Every interaction generates telemetry: clicks, filter changes, lesson starts, pronunciation attempts. reportActivityLog captures structured events with module, entity, action, element fields. This data feeds the AI personalization engine.

🔄 Dual-Mode Assessment

Exam Mode: 90-minute timer, hidden hints, real HKDSE simulation. Guide Mode: AI Coach available, hints enabled, learning-focused. Students choose their mode — building metacognitive awareness.

💡

Teacher Benefit: The analytics engine means you never teach in the dark. Instead of guessing who's struggling, you see it. Instead of assuming who's bored, you detect it. Data transforms teaching from reactive to proactive.

08

Gamification & Student Motivation

Progress psychology that keeps students coming back

🏆

Achievement System

4 tier visual levels: OOPS (0-50%), GOOD (51-60%), WELL DONE (71-90%), AMAZING (91-100%). Each tier triggers a different mascot expression.

OOPS 0-50%
GOOD 51-60%
WELL DONE 71-90%
AMAZING 91-100%
📊

Progress Visualization

Conic-gradient progress rings show completion at course level. The green-to-blue transition suggests growth becoming mastery. Streak indicators leverage loss aversion psychology — students don't want to break their chain.

🤖

ARI Mascot

The blue droplet character with thumbs-up gesture functions as a non-judgmental learning companion. Provides parasocial companionship during solo study. Non-authoritarian design normalizes mistakes.

🧠

The Science: Krashen's Affective Filter theory says anxiety blocks language acquisition. Every gamification element in Think & Speak exists to reduce that filter: friendly mascots (not judgmental teachers), encouraging messages (not red X marks), clear progression (not vague "study more"). The celebration effect on login streaks triggers dopaminergic rewards that reinforce habitual platform usage.

🎨 Tactile Design Psychology

Buttons use box-shadow: 0px 4px 0px 0px #bad6ee creating 3D press effects. The 2px downward translation on :active simulates physical button depression. This skeuomorphic design bridges physical toys and digital interfaces — making the learning platform feel immediately familiar to young learners who associate this feedback with fun.

09

Mobile-First Responsive Design

Learn anywhere — from classroom projector to commute smartphone

🖥️

Desktop

1920px+

  • 3-column grids
  • 40px titles
  • Full sidebar navigation
  • Teacher dashboard view
  • Sidebar + main content
📱

Tablet

768px–1919px

  • 2-column adaptive
  • 28px titles
  • Collapsible sidebar
  • BYOD classroom use
  • Touch-optimized
📲

Mobile

<768px

  • Single-column stack
  • 20px titles
  • Bottom-sheet modals
  • Commute study mode
  • Microlearning sessions
🔗

Cross-Device Continuity: Redux state persists via localStorage with debounced writes (500ms). Student starts a lesson on classroom desktop, continues on tablet at home, reviews on phone during commute — progress syncs seamlessly. The platform follows the student, not the other way around.

♿ Accessibility & WCAG 2.1 AA Compliance

High contrast ratios for visually impaired learners. Focus states with transition: all 0.2s ease-out. Semantic HTML structure supporting screen readers. prefers-reduced-motion media query respects user preferences. Inclusive design serves Primary through University age ranges in Hong Kong's diverse educational landscape.

10

Technical Architecture Overview

Enterprise-grade infrastructure designed for educational environments

⚛️ Frontend

React + Next.js + TypeScript

🔄 State

Redux Toolkit

🚪 Gateway

API + Auth

🧠 AI

Azure Speech + ML

💾 Data

Database + Analytics

⚛️ React + Next.js + TypeScript

  • Component-based architecture enabling modular development
  • CSS Modules with SCSS for scoped, maintainable styling
  • Server-side rendering for fast initial loads
  • Dynamic imports for code-splitting and lazy loading
  • Ant Design component library for consistent UI

🔄 Redux State Management

  • selectThinkAndSpeak and thinkAndSpeakActions patterns
  • Pagination persistence across browser refreshes
  • Optimistic UI updates with rollback on failure
  • Cross-module synchronization (Vocab ↔ Speaking ↔ Reading)

🧠 Azure Speech Integration

  • Real-time speech recognition at 16kHz sample rate
  • Phoneme-level analysis for pronunciation feedback
  • Translation between English, Mandarin, Cantonese
  • WebRTC/RecordRTC for audio capture

🔒 Security & Subscriptions

  • checkSubscriptionBeforeAction gating
  • Email verification first, then subscription check
  • JWT-based authentication with token refresh
  • Role-based access (Teacher/Student/Admin)
  • i18n: next-i18next for trilingual support
11

Reading Comprehension & DSE Prep

HKDSE-aligned exam simulation with AI Coach support

Dual-Mode Architecture

Exam Mode

90-minute countdown. Hints disabled. Real HKDSE pressure. Questions must be answered sequentially. Forced submission when time expires.

🤖

Guide Mode

AI Coach always available. Sentence-level hints. Strategy tips. Skip allowed. Learning-focused without time pressure. Build confidence gradually.

📋 9 Question Types (HKDSE Aligned)

  • True/False/Not Given — Distinguish false from unfound
  • Multiple Choice — Single and multi-select formats
  • Text Input — Free response with AI semantic evaluation
  • Table Completion — Structured data entry with fuzzy matching
  • Drag & Drop — Matching and classification exercises
  • Word Builder — Morphological awareness tasks
  • Referent Matching — Anaphora resolution (pronoun → referent)
  • Timeline Matching — Chronological ordering
  • Cloze Test — Contextual vocabulary filling

AI Coach Features

🎯 Passage Deconstruction

Students click any word, sentence, or paragraph to request AI analysis. Five knowledge domains:

  • Grammar & Sentence Structure
  • DSE Strategy & Exam Tips
  • Similar Questions & Practice
  • Vocabulary Deep Dive
  • Visuals & Multimedia Clips

Content prioritizes Chinese explanations for non-English locales, ensuring comprehension before translation.

📊 Part Result Analytics

After each part, students see: Overall score with color-coded icon (trophy 🏆 / rocket 🚀 / target 🎯 based on performance). Per-question breakdown with expandable feedback. Strategic debriefing with AI-generated tips. Next Steps calculation — system identifies which part to tackle next.

Exam Tip: The Reading Comprehension module maps directly to HKDSE Paper 1 formats. Students practicing here are training for the actual exam — not just generic reading exercises.

12

Course Adventure: Learning as a Journey

Structured paths that turn study into purposeful, motivating progression

🎬

Watch Video

Take Quiz

📖

Study Glossary

⬇️

Download PDF

🎤

Speaking Practice

How It Works

🗺️ Guided Learning Routes

Students never face a blank page. The Redux state machine tracks practicingCourseIdpracticingLessonIdpracticingTaskId. If a session is interrupted, students return to their last active checkpoint. The course grid uses gamified status badges (Last Practice, Completed, Premium) to surface relevant content.

🔒 Scaffolded Learning

isLessonLocked enforces prerequisite completion. Students must pass foundational vocabulary before accessing advanced speaking exercises. This scaffolding prevents overwhelm while giving advanced learners fast-track paths. Course list filtering enables teachers to assign specific courses to student groups.

For Teachers

📊

Course Management

Teachers can:

  • Assign specific courses to student subgroups
  • Track class completion rates by course
  • View individual student progress timelines
  • Export learning analytics for reports
  • Upload custom scripts for Read & Speak practice
🎯

Key Design: Course Adventure transforms curriculum delivery from teacher-driven distribution to AI-recommended personalized pathways. Each student's path is unique, but teachers maintain full visibility and control.

WORKSHOP SUMMARY

Think & Speak

Not a collection of tools — an interconnected environment where
assessment, practice, translation, and gamification reinforce each other.

🎯

Understand Each Student

Voice profiles reveal communication strengths. Identify who needs confidence vs. who's ready for challenges.

📊

Data-Driven Differentiation

Class analytics reveal invisible patterns. Spot struggling students early. Track intervention effectiveness in real-time.

🚀

Scaled Personalization

AI handles routine personalization. You focus on high-value human interactions that require your expertise.

Ready to explore?

Production: app.thinkandspeak.com
Teacher Account: test_teacher@seechange-edu.com / Password: Aa123456

Navigate slides  ·  N Speaker notes  ·  F Fullscreen  ·  Click cards for details

1 / 14

📝 Speaker Notes